From: Sean O'Dell <sean@...>
Date: 2003-06-20T02:27:31+09:00
Subject: Re: rb_protect and longjmp

"Paul Brannan" <pbrannan@atdesk.com> wrote in message
news:20030619170716.GQ1747@atdesk.com...
> On Thu, Jun 19, 2003 at 08:17:53AM +0900, Sean O'Dell wrote:
> > Looking for some insight about rb_protect and the longjmp calls.
> >
> > Is rb_protect supposed to always return back to the caller?  I'm
> > wrapping calls in rb_protect wherever I can in my C++ application, but
> > it seems that rb_protect doesn't always return.
>
> When rb_protect "doesn't return" what does happen?  Does your code get
> jumped over or does the program hang or does it crash?
>
> I know that if you throw a C++ exception from inside an rb_protect call,
> then on some platforms (g++ under Linux is one), std::terminate() will
> be called.

Well, now I'm trying to determine when it does what.  A simple test from the
top of the app works fine; rb_protect returns even though it called a
function that called rb_raise, but when I do the same thing from within a
ruby callback (my application embeds ruby as well as extends it, so this
ruby callback is actually a class method) rb_protect isn't protecting
anything; rb_protect never returns, as if I were calling rb_raise directly.
I think in one other place in my code where I did this, the program actually
just crashed.

I'm thinking there must be something goofy with the .dll build.  It works
trouble-free 98% of the time, but every so often something completely
unexpected happens like an invalid address access or something.  I'm
guessing I may need to double-check my compile options or something.

    Sean O'Dell